A unipolar-driven synaptic transistor for environment-adaptable vision system
Sukwon Jang, Keunho Soh, Chungryeol Lee, Taehyun Nam, Minjae Jang, Jeong-ik Park, Changhyeon Lee, Junhwan Choi, Jung Ho Yoon, Sung Gap Im
Abstract
Unlike conventional synapses in neural networks relying on bipolar spike-driven modulation, biological synapses in the peripheral nervous system handle unipolar input based on stimulus intensity, generating excitatory or inhibitory signals. Here, we demonstrate a unipolar-driven synaptic transistor (UDST) that exhibits both responses under a single-polarity voltage. The unipolar property is achieved within a single device through a bilayer gate dielectric, comprising a high-k charge trapping layer (k > 6) and an ultrathin charge tunneling layer (<5 nm), which synergistically facilitate dipole polarization and charge trapping. The UDST exhibits potentiation, depression, and adaptation while maintaining exceptional durability, with a dynamic range reduction of less than 0.9% over 2000 potentiation-depression cycles and minimal conductance variation of only 0.3%. This work presents the first implementation of a self-adaptive artificial vision system based on a unipolar-driven synaptic device, utilizing a 3 × 3 UDST array to achieve real-time object tracking and adaptive sensory processing without external control or computational peripheral circuits. The synapses in peripheral nerves of human eyes handle intensity-modulated unipolar input, different to the bipolar spike-driven modulation in artificial synapses. Jang et al. emulate this biological function in a single transistor device and implement it into a self-adaptive artificial vision system.